Characterisation of grunt sound pressure level from spawning Atlantic cod ( Gadus morhua )

Since sound from Atlantic cod ( Gadus morhua ) plays a significant role during their spawning activities, it is imperative to gain insights into their sound pressure levels. This knowledge is particularly valuable for understanding how cod responds to anthropogenic sounds, such as the intense sound from seismic air guns. In field experiments within sea cages using multiple instruments, including underwater cameras, a hydrophone, and an underwater acoustic vector sensor (AVS), the source level of cod grunts was scientifically measured. The measurements yielded an estimated median source level of 123.1 dB re 1 µ Pa at 1 m (121.8–124.2 dB, 95% confidence interval). The compact AVS measured collocated acoustic pressure and acoustic particle motions in three dimensions, providing valuable information regarding acoustic directionality in addition to sound pressure. This directional information assists in detecting the grunt bearing, estimating probable grunt propagation ranges, and subsequently reducing uncertainties associated with the estimation of the grunt source level. These experiments have demonstrated the AVS measured the cod vocalisations with directional information and can be potentially broadly used in other marine bioacoustics monitoring and tracking research.


Introduction
Fish produce sound for communication, attracting partners and defending their territory (Hawkins and Myrberg 1983;Fay and Popper 2000;Hawkins and Picciulin 2019;Hawkins and Popper 2020).Among fish families, Gadoids (cod fish) are known for being some of the most vocal (Hawkins and Picciulin 2019).Gadoids produce sounds and hear the best in the frequency range 50-500 Hz (Hawkins and Picciulin 2019;Hawkins and Popper 2020).Fish mainly sense sound through particle motion, although some groups of fish also detect sound pressure (Nedelec et al. 2016;Popper and Hawkins 2018;Putland et al. 2018).The link between particle motion in sound waves and the behavioural responses of marine organisms has remained elusive due to practical limitations (Nedelec et al. 2016), although some descriptions of particle motion for anthropogenic sounds and natural soundscapes are now becoming available (Hawkins and Popper 2020;Jones et al. 2022).
One of the most extensively studied Gadoids, Atlantic cod (Gadus mohua), produce sounds during both adult and juvenile phases (Brawn 1961), particularly during spawning, courtship, and agonistic behaviour (Rowe and Hutchings 2006).These sounds can be used to locate spawning aggregations (Hawkins 2022;Van Hoeck et al. 2023).Although cod produce various sounds, the grunt is the most well described and is used both during and outside the spawning period (Rowe and Hutchings 2006;Laurijssen et al. 2022;Seri et al. 2023).To understand how anthropogenic sound may affect cod behaviour and vocalisation, it is crucial to have a detailed description of their sound characteristics, in particular the source level, and understand particle motion from the fish perspective (Hawkins and Popper 2020;Soudijn et al. 2020).
Noise can induce stress and reduce the egg production and fertilisation rate of spawning cod (Sierra-Flores et al. 2015).Cod were also seen to leave the area during a seismic survey outside the spawning season (Engås et al. 2011).However, seismic air gun exposure did not result in the spawning cod abandoning their spawning habitat (McQueen et al. 2022(McQueen et al. , 2023) ) Sound from seismic airguns has been reported to affect swimming depth (Davidsen et al. 2019;McQueen et al. 2022) and activity level in cod (van der Knaap et al. 2021).Furthermore, noise may also impact cod at the larva stage in terms of reduced growth (Nedelec et al. 2015) and orientation (Cresci et al. 2023).
Studies have also shown that vessel traffic noise could significantly reduce the communication range for cod grunts (Stanley et al. 2017).To fully understand the impact of noise on communication, it is necessary to determine the amplitude of cod vocalisation, specifically its source level (SL).Unfortunately, knowledge on this topic is limited.In field measurements, uncertainties in estimating SL increase with unknown ranges between the cod and the measuring instruments, such as a hydrophone.For instance, a grunt SL of 127 dB re 1 μPa at 1 m was observed from a single cod kept in the tank (Nordeide and Kjellsby 1999), while a mean SL of 163.5 ± 7.9 dB re 1 μPa at 1 m was estimated through a survey in Austevoll, Norway (Seri et al. 2023).The difference of 36 ± 7.9 dB between these measurements highlights the significant uncertainties associated with cod grunt SL.
A SL of 127 dB re 1 μPa at 1 m was used to estimate the impact of boat noise on the effective communication distance of cod (Stanley et al. 2017).A more precise estimate of SL would greatly enhance the possibilities for predicting the impact of different types of noise on cod communication.Moreover, masking is not simply a result of comparison of sound levels.Fish and cod in particular have been shown to hear directionally due to their sensitivity to particle motion (Buwalda et al. 1983).Directional hearing could increase the efficacy of communication through a process called spatial release from masking, which means that animals may discriminate informative from unwanted noise if the source is located in a different direction than the masking source (Erbe et al. 2016).To understand this process, it is crucial to describe both the cod vocalisations and the noise sources in terms of particle motion.To our knowledge, we provide a first description of the cod grunt directionality using acoustic particle motion.
Field experiments were conducted to measure the cod vocalisations, from 9 March 2020 to 11 March 2020, using four sea cages of wild adult cod.Each cage was equipped with underwater cameras to monitor the cod behaviour, and their vocalisation was recorded by a data acquisition system using a hydrophone and an underwater acoustic vector sensor (AVS).The recorded grunts from each cage were synchronised with those captured by the bottom camera, the hydrophone and the AVS, after which they were separated and subjected to analysis.

Experiment, instrumentation, and data collection
The field experiment was conducted from 9 March 2020 to 11 March 2020, at the Institute of Marine Research's (IMR's) Austevoll Research Station in Austevoll, Norway (Figure 1(a)).There were four sea cages (C1-C4) of size 5 × 5 × 5 m 3 with wild adult cod that were caught around Austevoll islands, and the cod distributions in each cage was between 18 and 20 individuals (Table 1).The ratio between male and female cod was approximately 1:1, which was similar among cage C1-C4.The four cages were installed close to the shore at a depth of 8 m, and two cameras from GoPro Inc. (model: Hero 7 Black) were used in each cage.The cameras were kept in a waterproof enclosure using external power.In each cage, one camera pointed up, the other pointed down, and both recorded video with sound.
An external data acquisition system was built using National Instrument Corporation's PXI system, it synchronously digitised the input analogue signals from the B&K and the AVS continuously at a sample rate of 20 kHz, and saved files every 10 s for post processing.In the following analysis, without losing integrity, the raw data were initially decimated to a sample rate of 2 kHz (low pass filtering first and then down sampling to 2 kHz) and then analysed with reduced computational load.
The measurements were conducted as part of the project 'SpawnSeis', which investigated how seismic air guns affected cod behaviour.During the 3 days of data collection, active air gun exposure occurred for 3 hours each day.These 3 hours were excluded from the analysis.Fish might change their vocalisation in response to anthropogenic noise, including adjustments in timing, amplitude, or frequency (Radford et al. 2014).However, based on observations that the cod conducted similar behaviour as before shooting started in this period, it was unlikely that such changes would occur outside of the actual exposure period.

Signal processing methods
The AVS measures sound pressure p t ð Þ (channel O) and orthogonal particle velocities, specifically Y and Z).The direction of arrival can be directly measured using the active acoustic intensity vectors (Felisberto et al. 2010;Tesei et al. 2019;Thode et al. 2019;Zhang et al. 2023).The active intensity is parallel to the local net transport of acoustic energy.For each time-frequency, the active intensity vectors are defined as , the symbol '*' indicates complex-phase conjugation, and Re½� represents the real part operator for a complex number.Assuming channel X points up, azimuth θ f ; t ð Þand elevation φ f ; t ð Þcan be measured by inverting the tangent functions of the active intensity as Since I c f ; t ð Þ can be computed similarly to a conventional spectrogram using fast Fourier transform, θ f ; t ð Þ is named as 'azigram' (Thode et al. 2019), and hereby φ f ; t ð Þ is analogous termed as 'elegram' in this context.The azimuth (0-360°) is defined as the angle of the incoming acoustic wave when channel Y is pointing to a fixed direction and calculated clockwise from channel Y.The elevation (0-180°) is defined as the vertical angle of the incoming acoustic wave when channel X points up, where 0° represents upwards and 180° corresponds to downwards.Following the guidelines as best practice for in-situ measurement of underwater sound (Robinson et al. 2014), all the parameters used for data processing are listed in Table S1 of Supplementary Material, which presents the verification of the bearing measurements through a trial with an active 400 Hz active source transmitting from a known position relative to the AVS.
Provided by the grunt directional information, it was possible to determine the direction from which the cod vocalised.A data masking technique was used to extract time-frequency elevations from the elegram.First, a mask matrix was prepared using the spectrogram of a known grunt.The mask had a time length of 500 ms which corresponds to five segments of the grunt spectrogram, and it encompassed the grunt's frequency components.Second, only the matrix element indexes were kept valid, where the power spectral density values must exceed a certain threshold.The optimal threshold for each grunt's elevation extraction depended on the respective grunt SNR.For simplicity based on visual observations, a threshold of 80 dB was used as follows.The mask matrix retained only the element positions which the density values were over 80 dB.Subsequently, this matrix was used to obtain the elevations from the respective elegram.Note that the threshold of 80 dB was chosen as a simplification, and its optimality for all grunts was not validated.Frequency components around 50 Hz and the harmonics of 100, 150, 200 and 250 Hz were removed from the analyses since this was noise not related to cod sound.The mask covered the grunt's frequency components, which were divided into four frequency ranges: 52-98 Hz, 102-148 Hz, 152-198 Hz and 202-248 Hz.Once the time-frequency elevations are obtained for a grunt, the median value was used as this grunt elevation.

Data scrutiny
The cod grunts were manually selected to ensure grunt data quality.It was crucial to distinguish the cod grunts originating from each cage, as both the B&K hydrophone and the AVS received grunts from cage C1-C4.The underwater cameras captured audio recordings that included grunts, most likely recorded when the cod was at the front of the camera.The specific characteristics of the directional microphone of the cameras were unknown, especially within a waterproof encapsulation underwater, but it was probable that the microphone had low sensitivity.For instance, through data scrutiny, it was found out that the grunts recorded by the camera on the bottom of cage C1 could not be identified in the audio recordings of the camera on the surface of cage C1.Therefore, it was assumed that the camera microphones could only record the cod grunts from the same cage as the camera.By taking advantage of the bottom camera in each cage, the grunts recorded by the camera were used to match and identify the respective grunts received by the B&K hydrophone and the AVS.Subsequently, this allowed for the separation of grunt samples from each cage as the camera.
Grunt samples from each cage were carefully chosen after visual inspection (using the audio editing program Cool Edit Pro) based on our experience of grunt waveform and spectrogram.Each grunt was manually extracted from the acoustic data in the time domain, with the beginning of a grunt marked as started when the first grunt pulse emerged above background noise, and the end noted when the last grunt pulse subsided and blended back into the background noise.The uncertainties in grunt length were neglected in the extraction process, as they were significantly smaller compared to a fixed grunt time length.Due to overlap in a spectrogram, it is inaccurate to select a grunt according to the time length on the spectrogram.Additionally, selecting the grunt samples that arrived at similar time was avoided.For the analysis, a total of 158, 163, 122, and 129 grunts were selected from cage C1-C4, respectively.
The recorded grunts contained background noise, particularly ambient noise and instrumentation noise.It is essential to minimise the noise impact when calculating the received grunt sound pressure level (SPL).In Figure 3, the frequency characteristics of the three grunts are not identical, and the background noise frequency characteristics are different.Comparing with the B&K, the AVS experienced stronger background noise (<30 Hz) and electric noise (harmonics of 50 Hz in Figure 3(b)).For both the B&K and the AVS, a bandpass filter (30-500 Hz) was first applied to exclude noise from frequencies below 30 Hz, before calculating the received grunt SPL.When selecting a grunt, a short period of noise ahead or after this grunt was also manually chosen to obtain the background noise power.Thereafter, the noise power within the bandwidth (30-500 Hz) was subtracted when calculating the SPL.The grunt and noise (from the B&K and the AVS) were always selected synchronously when the B&K was used as the reference due to its lower background noise characteristics.
A grunt SL was calculated as dB re1μPa at1m.The background noise at a receiving hydrophone is always included in receiving a grunt, and hence the noise level (NL) must be first removed from the received grunt SPL (dB re1μPa) before calculating the grunt SL.For the m th received grunt with a received SPL RL m ð Þ, when the cod stayed in a range r to a hydrophone, the NL NL m ð Þ is initially removed from RL m ð Þ, the grunt transmission loss (TL) TL m ð Þ is compensated, and then SL m ð Þ is estimated by where g RMS m ð Þ and n RMS m ð Þ are the root mean square (RMS) values of the received grunt and the respective noise, and TL m ð Þ is simplified as the spherical spreading (Urick 1983).When r is unknown, especially in field measurements, there are significant challenges to compensate the TL.As a measure of the desired grunt strength to the noise, the grunt signal-to-noise (SNR) ratio is defined as This ratio indicates the quality of a grunt sample.S1 of Supplementary Material.

Grunt time length
The cod grunts identified had varying durations (see the examples in Figure 4).Among the five grunt examples, the second grunt is the longest with a duration of 500 ms, which is five times longer than the shortest grunt with a duration of 100 ms.Therefore, it is wrong to calculate a grunt SL assuming all the grunts have the same duration.It is likely that the male cod who vocalise during spawning (Brawn 1961), and there was no observable relationship between grunt duration and male cod length (Figure 5).Although the cod grunts analysed had varying durations, similar median values were observed, particularly 173 ms, 183 ms, 186 ms, and 183 ms for cages C1-C4, respectively.The statistics for grunt duration reveal similar 25th percentile and 75th percentile values, as well as median values (Table 2).The 95% confidence intervals round the median values of cage C1-4 are 165-180 ms, 178-188 ms, 178-194 ms, and 178-188 ms, respectively.This indicates that most of the grunts have a time length of within 200 ms.

Grunt spectrum
The power spectrum of a grunt provides insight into the relative magnitudes of its frequency components.The grunts recorded by the B&K hydrophone contained much less 50 Hz harmonics noise (see Figure 3) and hence they were used to observe the grunt power spectra (Figure 6).The dominant frequency components vary among the grunts (Figure 6  visualises the dominant frequency components presented in all the spectra, and it confirms the presence of three dominant frequency ranges.

Grunt directionality
The incoming sound directionality to the AVS was determined by using the method in Section 2.2.The cod grunts' directionality (azimuth and elevation) was displayed together with the spectrogram (Figure 7).As highlighted by the white arrows, two groups of grunts arrived around 00:05:28 and 00:06:15 (UTC time) and the spectral density is visualised on the spectrogram (Figure 7(a)), and the respective arriving azimuths and elevations at the AVS are displayed (Figure 7(b-c)).These two groups of grunts originated from two different azimuths, shown by the orange and blue colours, and their elevations were notably distinct, as displayed by the orange and blue colours.
An elevation of 90° corresponds to the horizontal plane at the same depth as the AVS.If a grunt's elevation is less than 90°, it indicates that the grunt originates from a fish above the AVS.Conversely, if a grunt's elevation is greater than 90°, it suggests that the grunt originates from a fish below the AVS.The grunt elevation in cage C1 varied from 33° to 142°, with a concentration around the median value of 79° (Figure 8(a)).Most  grunts had an SNR over 5 dB with only one grunt SNR of −2.7 dB (Figure 8(b)), which indicates strong grunt signals relative to the background noise that ensure the AVS resolves the grunt directionality.From the elevation, most grunts were made when the cod were at a similar depth to the AVS, while a few grunts were produced above and below the AVS.Possible cod depths (1.47-2.30m) were numerically calculated given by the elevations 77-81° (95% confidence interval).After reviewing the bottom camera  videos (at the grunt time when it was bright enough to observe), it was evident that most of the fish were in the middle of the water column.

Grunt source level
The received grunt SPL varied with the cage (Figure 9), as the distances between the cages and the hydrophones were different.The median SPL values for both the B&K and the AVS decreased with the range (range from a cage to the B&K and the AVS).As the furthest cage, cage C3 had the lowest median SPL value, when the SPLs of the AVS exhibited more uncertainty due to low SNR, with a median value of −19 dB.Conversely, cage C1 had high SNRs, particularly there was only one grunt SNR of −2.7 dB and the 25th percentile value of the SNR was 9.4 dB for the AVS.Compared with the B&K, the SPL distribution of the AVS resembled a shift.Due to the high SNR, the grunts from cage C1 were used to estimate the grunt SLs.
Given by the numerical simulation of the possible cod positions according to the grunt elevations (77-81°), the median range values to the B&K and the AVS were calculated as 1.71 m and 1.97 m, which were used to compensate TL in Equation (4).Table 3 shows the estimated SL statistics from the B&K and the AVS, in which the differences in the 25th percentile, 75th percentile and median values are 1.3 dB, 1.2 dB, and 1.1 dB, respectively.Considering the uncertainties in calibration and the simplification in TL compensation, it could be concluded that the grunt SL statistics from the B&K and the AVS were similar.Since the B&K was calibrated by IMR at the trial start, the SL values from the B&K were much preferred hereafter, particularly a median SL of 123.1 dB re μPa at 1 m with the 95% confidence interval 121.8-124.2dB re μPa at 1 m.

Discussion
Field experiments were conducted to characterise SPLs of grunts produced by spawning Atlantic cod using various instruments.The underwater cameras helped identify grunts originating from each of the four sea cages, allowing for the estimation of grunt SLs from a specific cage.Estimating the grunt SL is particularly challenging in field measurements because the grunt propagation range is often unknown (Stanley et al. 2017;Seri et al. 2023), which increases the uncertainty.In this study, the dimensions of cage C1 limited the maximum propagation range, and this approach reduced uncertainty in the measurements compared to those previously published.Consequently, the estimated grunt SLs had lower uncertainty, with a median value of about 123.1 dB re 1 μPa at 1 m (95% confidence interval: 121.8-124.2dB).For future research, it is promising to apply video and audio arrays (Mouy et al. 2023) for estimating the SL of each individual grunt, when each grunt travel distance could be even more precisely determined.Table 3. Source level for the cod grunts after compensating the averaged transmission range: statistics of SL, NL, and SNR (158 grunts from cage C1).TL is compensated using Equation (4), using by the ranges 1.27 m and 1.46 m to the B&K and the AVS, respectively.SNR is calculated using Equation ( 6).The 95% confidence intervals around the median values are 121.8-124.2dB for the B&K and 121.0-122.7 dB for the AVS.Several studies have described the vocalisations of spawning Atlantic cod (Brawn 1961;Rowe and Hutchings 2006), particularly the cod grunts (Wilson et al. 2014).These studies found similar frequency distributions with several frequency peaks, resulting from the pulsed structure of the grunt (Hawkins and Rasmussen 1978).However, only a few reported grunt SL, with large differences between reported values: 127 dB re 1 µPa at 1 m (Nordeide and Kjellsby 1999), 120-133 dB re 1 µPa at 1 m (Midling et al. 2002), and 163 dB re 1 µPa at 1 m (Seri et al. 2023).Accurate measurement of the cod grunt SL is crucial to understand how vocalisations from spawning cod may be impacted by anthropogenic noise, such as shipping noise, particularly through noise masking and reduction of effective communication range (Stanley et al. 2017;Putland et al. 2018), and this requires knowledge of the grunt SL for calculations.When calculating the reduction in communication range, Stanley et al. (2017) used the reported grunt SL of 127 dB re 1 µPa at 1 m from Nordeide and Kjellsby (1999).With a simple spherical spreading transmission loss (assuming sound spreads evenly in all directions), the grunt would reach a hypothetical background noise level of 100 dB re 1 µPa after about 20 m.However, if using the grunt SL of 163 dB re 1 µPa at 1 m reported by Seri et al. (2023), the grunt could propagate over more than 1 km before reaching the same background noise level.The large difference underscores the significance of having knowledge about the sound SL of fish vocalisations.
Spawning sounds may vary between populations (Rowe and Hutchings 2006).The previously reported SL studies involved coastal cod in Norway, with the study by Seri et al. (2023) conducted in the same area as the experiment in this study, hence population differences may not explain the large discrepancy.Grunt differences may also be related to the size of the drumming muscle (Rowe and Hutchings 2008), but this seems unlikely to explain such a large difference as the fish were from the same population, with a range of size distributions.However, Nordeide and Kjellsby (1999) found elevated sound levels within the frequency of cod grunts on hydrophones in a Norwegian fjord far away from the closest locations where cod had been observed.The received sound levels were much higher than expected at such a distance from the source level calculated from an aquarium recorded cod.They posed the hypotheses that the summation of many (e.g.1000 grunts) overlapping cod grunts in a chorus could lead to higher sound levels, and thus, propagation distances.An alternative or additional hypothesis would be that the grunts become louder in the field, where they would have to bridge larger distances to communicate, than when they are constrained in a tank or a sea cage.Cod have been shown to use a larger repertoire of calls in the field compared to the laboratory, but it is unclear whether sound levels can differ (Midling et al. 2002).These hypotheses may explain the differences in sound level between Seri et al. (2023) and our study.
This study used the active intensity method to extract grunt directionality from the acoustic particle motion measured by the AVS, which is particularly relevant to exploit information of particle motion, as fish perceive sound as directional (Popper and Hawkins 2018).It is demonstrated that the AVS has the potential to be a valuable instrument for monitoring fish sound production and behavioural responses to sound stimuli.Furthermore, the elevation information assisted in estimating fish ranges to calculate the cod grunt SLs.Comparing the AVS hydrophone with the B&K SL measurements, similar grunt SL statistics were obtained from the two different instruments, ensuring the validity of the measurements.
It is necessary to avoid errors by quality assurance, where data scrutiny is highly recommended when calculating a received grunt SPL.A critical factor was noise contamination from the instrumentation.For instance, the AVS exhibited a noise level approximately 20 dB higher than the B&K, which potentially could have resulted in overestimated SPLs if the noise effect was not removed.Additionally, since the grunt duration varies (Finstad and Nordeide 2004), inaccurate SPLs could be obtained if using a fixed time length for all grunts.A longer grunt duration would have resulted in a lower SPL, and vice versa.Therefore, manual grunt selection with data scrutiny was adopted to ensure that a valid duration was used for each grunt.
Due to limitations in our experiment, we were unable to assign grunts to specific individuals.Consequently, the grunt SPL on individual fish and its variations with factors like size are unavailable.It is also possible that one or a few fish conducted most of the vocalisations.Future studies should target identification of variations in individual vocal performance.Additionally, investigations are needed to determine whether cod vocalisations exhibit changes in SPL during the spawning season, or in response to noise conditions, such as anthropogenic noise.Understanding SPL adjustments in noisy environment is particularly important to assess if cod can compensate for reduced communication range by increasing their sound level.

Figure 1 .
Figure 1.Setup of the experiments measuring the cod sound.(a) Experiment location: The measurements were conducted at the Austevoll Research Station, Institute of Marine Research, Norway.The red star denotes the AVS location.The orange dash lines assist visualising the four quadrants of the AVS directional measurements.(b) Overview of four sea cages C1-C4 and the instrument deployment in cage C1: The AVS was deployed with channel X measuring the vertical acceleration component (pointing upward) and channel Y and Z measuring horizontal acceleration components (channel Z pointing towards the station).(c) Instrument setup in cage C1: The orange dot denotes the AVS.Two cameras GoPro a and b (model: Hero 7 Black) were used to monitor the cod behaviour.One camera was positioned close to the surface facing down and the other was placed near the bottom of the cage facing up.On the right side, there are images of the camera with a waterproof enclosure (top image) and the AVS with mount rack (bottom image).

Figure 3 .
Figure 3. Spectrogram of three cod grunts from cage C1 on 9 March 2023.(a) The B&K hydrophone.(b) The AVS channel O.The x-axis is UTC time (HHMMSS).The processing parameters are shown in TableS1of Supplementary Material.
(a)), particularly the maximum magnitudes of the grunts.As highlighted by black arrows, three dominant frequency ranges (P1-P3 in Figure 6(a)) are observed, specifically 44-68 Hz, 93-145 Hz and 185-220 Hz.It is apparent that the range P2 is the most frequently occurring component.The lowest dominant grunt frequency was 44 Hz (the red arrow).In Figure 6(b), the summation of the power spectra of 573 grunts

Figure 4 .
Figure 4. Examples of five cod grunts with different amplitudes from cage C1.

Figure 5 .
Figure 5. Overview of the grunt time length and the male cod length.The boxplots show minimum, 25th percentile, median, 75th percentile and maximum values.The boxplots in blue are grunt time length, and those in cyan with hatches are male cod length.The notches indicate the 95% confidence interval around the median values.

Figure 6 .
Figure 6.Overview of grunt power spectra (without 30-500 Hz bandpass filtering).(a) Power spectra of 573 grunts from cage C1-C4, recorded by the B&K hydrophone.The magnitude of each power spectrum is normalized to 1. P1-P3 highlight the frequency ranges 44-68 Hz, 93-145 Hz and 185-220 Hz.The red arrow highlights the lowest dominant frequency, 44 Hz.The frequency components < 25 Hz are the background noise.The number of fast Fourier Transform was 2000, and rectangular window was used.(b) Summation of the magnitudes of 573 power spectra in (a).The black arrows highlight three frequency ranges, denoted by P1-P3.

Figure 7 .
Figure 7. Spectrogram, azigram and elegram from 00:01:31 to 00:09:25 UTC time, 9 March 2023.(a) Spectrogram including cod grunts highlighted by the white arrows.(b) Azigram shows the cod grunts came from different azimuths to the AVS, as displayed by the orange and blue colours.(c) Elegram shows the cod grunts came from below and above the AVS, as displayed by the blue and orange colours.The x-axis is UTC time (HHMMSS).The processing parameters are shown in Table S1 of Supplementary Material.The arrows highlight the grunt examples.

Figure 8 .
Figure 8. Overview of grunt elevations and SNR of 158 grunts from cage C1.(a) Elevation of each grunt.Each elevation is the median value of all the valid masked time-frequency elevations.The green, blue, and black lines show the 25th percentile, median, and 75th percentile values, particularly 73°, 79°, and 84°.(b) SNR of each grunt, measured by the AVS channel O.

Figure 9 .
Figure 9. Overview of the received grunt SPL (blue boxplots), NL (black boxplots) and SNR (cyan boxplots with hatches), when the grunt propagation loss is not compensated.NL is removed when calculating the SPL.The boxplots are presented in the logarithmic scale, with minimum, 25th percentile, median, 75th percentile and maximum values.The notches represent the 95% confidence interval values around the median.

Table 1 .
Cod distribution in each cage.
Figure 2. Sensitivity of the AVS directional channel X, Y and Z.

Table 2 .
Statistics of cod grunt time length of cage C1-C4.